Methods to Expand Cell Signaling Models Using Automated Reading 155Figure 5 summarizes results of our extension methods on 1232 interactions
with respect to new node connections to the model:(i) number of new nodes regulating baseline model elements, not regulated by
baseline model elements (dark blue);
(ii) number of new nodes regulating baseline model elements, not regulated by
any element, baseline or new (red);
(iii) number of new nodes regulated by baseline model elements, not regulating
any elements in the baseline model (yellow);
(iv) number of new nodes regulated by baseline model elements, not regulating
any element, baseline or new (purple);
(v) number of new nodes inserted into existing pathway - new regulators of
baseline model elements that are also regulated by baseline model elements
(green);
(vi) number of new nodes as intermediate elements of new pathways when mul-
tiple extensions are connected (light blue);
(vii) total number of all elements used in the extension method (dark red).
In Fig. 6 (a), four different sections can be observed, and each section corre-
sponds to one of the extension methods. Each method has its unique feature. For
example, the ND method only includes relationships relevant to one layer, and
this makes the number of new elements added to the model significantly smaller
than other methods. Also, the light blue nodes indicate the number of newly
added elements that are in a newly formed pathway. Since CD and ND do not
include indirect parent interactions, we can see that the number of elements in
new pathway is 0. While in CI and NI, we can tell that indirect interactions are
included. The numbers within one method show higher similarity, but we can
still observe some patterns. For example, the cumulative parent-set methods,
CD and CI show an increase in the number of new nodes when more layers are
considered. Furthermore, since NI has cumulative parents when they finish the
noncumulative part, they also experience an increase when the step of noncu-
mulative part is fixed. The numbers saturate at around 600, which is due to the
limited size of baseline model and extensions we have. This is also the reason we
choose to perform the cumulative approach for at most 3 steps.
In general, choosing the method to extend the model depends on the scenario
a user is interested in. For example, if the focus is on the regulation of a specific
element, one can track down each layer of parents using ND, and see the change
of the model after modifying that specific layer. On the other hand, if the goal
is to include as many new stimuli as possible with a fewer number of layers,
cumulative methods such as CI or CD will fit better. We selected 20 elements
as part of the base layer, since these elements appear in properties that we are
testing, leading to relatively large base layer given the size of the baseline model.
Therefore, by incorporating elements related to more than one layer, we capture
almost all extensions related to the baseline model. Thus, the ‘All In’ method,
which adds all extension interactions to the baseline model at once, does not
change the counts shown in Fig. 6 (a), when compared to many cases of CD, CI
and NI methods.